fs2: fix cgroup.subtree_control EPERM on rootless + add CI

Fix #2339 (regression in 813cb3e)

Signed-off-by: Akihiro Suda <[email protected]>

.travis.yml

@@ -38,6 +38,8 @@ matrix:
         - sudo ssh default -t 'cd /vagrant && sudo make localintegration RUNC_USE_SYSTEMD=yes'
         # same setup but with fs2 driver instead of systemd
         - sudo ssh default -t 'cd /vagrant && sudo make localintegration'
+        # rootless
+        - sudo ssh default -t 'cd /vagrant && sudo make localrootlessintegration'
   allow_failures:
     - go: tip
 

Vagrantfile

@@ -19,5 +19,7 @@ install podman make golang-go libseccomp-devel bats jq
 ts run
 EOF
     dnf clean all
+    # Add a user for rootless tests
+    useradd -u2000 -m -d/home/rootless -s/bin/bash rootless
   SHELL
 end

libcontainer/cgroups/fs2/create.go

@@ -65,7 +65,7 @@ func neededControllers(cgroup *configs.Cgroup) ([]string, error) {
 
 // CreateCgroupPath creates cgroupv2 path, enabling all the
 // needed controllers in the process.
-func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
+func CreateCgroupPath(path string, c *configs.Cgroup, rootless bool) (Err error) {
 	if !strings.HasPrefix(path, UnifiedMountpoint) {
 		return fmt.Errorf("invalid cgroup path %s", path)
 	}
@@ -79,6 +79,7 @@ func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
 	elements := strings.Split(path, "/")
 	elements = elements[3:]
 	current := "/sys/fs"
+elementLoop:
 	for i, e := range elements {
 		current = filepath.Join(current, e)
 		if i > 0 {
@@ -95,13 +96,25 @@ func CreateCgroupPath(path string, c *configs.Cgroup) (Err error) {
 					}
 				}()
 			}
+			// Write cgroup.type explicitly.
+			// Otherwise ENOTSUP may happen.
+			cgType := filepath.Join(current, "cgroup.type")
+			if err := ioutil.WriteFile(cgType, []byte("threaded"), 0755); err != nil && !rootless {
+				return err
+			}
 		}
 		// enable needed controllers
 		if i < len(elements)-1 {
 			file := filepath.Join(current, "cgroup.subtree_control")
 			if err := ioutil.WriteFile(file, []byte(allCtrs), 0755); err != nil {
-				// XXX: we can enable _some_ controllers doing it one-by one
-				// instead of erroring out -- does it makes sense to do so?
+				if rootless {
+					// try write one by one
+					for _, ctr := range ctrs {
+						_ = ioutil.WriteFile(file, []byte(ctr), 0755)
+					}
+					// ignore unwritable files
+					continue elementLoop
+				}
 				return err
 			}
 		}

libcontainer/cgroups/fs2/fs2.go

@@ -53,7 +53,10 @@ func (m *manager) getControllers() error {
 
 	file := filepath.Join(m.dirPath, "cgroup.controllers")
 	data, err := ioutil.ReadFile(file)
-	if err != nil && !m.rootless {
+	if err != nil {
+		if m.rootless && m.config.Path == "" {
+			return nil
+		}
 		return err
 	}
 	fields := strings.Fields(string(data))
@@ -66,10 +69,23 @@ func (m *manager) getControllers() error {
 }
 
 func (m *manager) Apply(pid int) error {
-	if err := CreateCgroupPath(m.dirPath, m.config); err != nil {
+	if err := CreateCgroupPath(m.dirPath, m.config, m.rootless); err != nil {
+		// Related tests:
+		// - "runc create (no limits + no cgrouppath + no permission) succeeds"
+		// - "runc create (rootless + no limits + cgrouppath + no permission) fails with permission error"
+		// - "runc create (rootless + limits + no cgrouppath + no permission) fails with informative error"
+		if m.rootless {
+			if m.config.Path == "" {
+				cl, clErr := neededControllers(m.config)
+				if clErr == nil && len(cl) == 0 {
+					return nil
+				}
+				return errors.Wrap(err, "rootless needs no limits + no cgrouppath when no permission is granted for cgroups")
+			}
+		}
 		return err
 	}
-	if err := cgroups.WriteCgroupProc(m.dirPath, pid); err != nil && !m.rootless {
+	if err := cgroups.WriteCgroupProc(m.dirPath, pid); err != nil {
 		return err
 	}
 	return nil
@@ -199,7 +215,11 @@ func (m *manager) Set(container *configs.Config) error {
 		}
 	}
 	// devices (since kernel 4.15, pseudo-controller)
-	if err := setDevices(m.dirPath, container.Cgroups); err != nil {
+	//
+	// When m.Rootless is true, errors from the device subsystem are ignored because it is really not expected to work.
+	// However, errors from other subsystems are not ignored.
+	// see @test "runc create (rootless + limits + no cgrouppath + no permission) fails with informative error"
+	if err := setDevices(m.dirPath, container.Cgroups); err != nil && !m.rootless {
 		errs = append(errs, err)
 	}
 	// cpuset (since kernel 5.0)
@@ -218,7 +238,7 @@ func (m *manager) Set(container *configs.Config) error {
 	if err := setFreezer(m.dirPath, container.Cgroups.Freezer); err != nil {
 		errs = append(errs, err)
 	}
-	if len(errs) > 0 && !m.rootless {
+	if len(errs) > 0 {
 		return errors.Errorf("error while setting cgroup v2: %+v", errs)
 	}
 	m.config = container.Cgroups

libcontainer/cgroups/systemd/v2.go

@@ -151,7 +151,7 @@ func (m *unifiedManager) Apply(pid int) error {
 	if err != nil {
 		return err
 	}
-	if err := fs2.CreateCgroupPath(m.path, m.cgroups); err != nil {
+	if err := fs2.CreateCgroupPath(m.path, m.cgroups, m.rootless); err != nil {
 		return err
 	}
 	return nil

tests/integration/cgroups.bats

@@ -95,7 +95,7 @@ EOF
 
     runc run -d --console-socket $CONSOLE_SOCKET test_cgroups_permissions
     [ "$status" -eq 1 ]
-    [[ ${lines[1]} == *"cannot set pids limit: container could not join or create cgroup"* ]]
+    [[ ${lines[1]} == *"rootless needs no limits + no cgrouppath when no permission is granted for cgroups"* ]] || [[ ${lines[1]} == *"cannot set pids limit: container could not join or create cgroup"* ]]
 }
 
 @test "runc create (limits + cgrouppath + permission on the cgroup dir) succeeds" {

tests/rootless.sh

@@ -82,6 +82,24 @@ function enable_cgroup() {
 		# handling.
 		[[ "$cg" == "cpuset" ]] && chown rootless:rootless "$CGROUP_MOUNT/$cg$CGROUP_PATH/cpuset."{cpus,mems}
 	done
+    # cgroup v2
+    if [[ -e "$CGROUP_MOUNT/cgroup.controllers" ]]; then
+        # Enable controllers.
+        for f in $(cat "$CGROUP_MOUNT/cgroup.controllers"); do echo +$f > "$CGROUP_MOUNT/cgroup.subtree_control"; done
+        # Create the cgroup.
+        mkdir -p "$CGROUP_MOUNT/$CGROUP_PATH"
+        # chown/chmod all cgroup.procs.
+        # See https://www.kernel.org/doc/html/latest/admin-guide/cgroup-v2.html#delegation-containment
+        chown root:rootless "$CGROUP_MOUNT/$CGROUP_PATH" "$CGROUP_MOUNT/$CGROUP_PATH/cgroup.procs" "$CGROUP_MOUNT/cgroup.procs"
+        chmod g+rwx "$CGROUP_MOUNT/$CGROUP_PATH" "$CGROUP_MOUNT/$CGROUP_PATH/cgroup.procs" "$CGROUP_MOUNT/cgroup.procs"
+        # Fix up cgroup.type.
+        echo threaded > "$CGROUP_MOUNT/$CGROUP_PATH/cgroup.type"
+        # Make sure cgroup.type doesn't contain "invalid". Otherwise write ops will fail with ENOTSUP.
+        # See http://man7.org/linux/man-pages/man7/cgroups.7.html
+        if grep invalid "$CGROUP_MOUNT/$CGROUP_PATH/cgroup.type"; then
+            exit 1
+        fi
+    fi
 }
 
 function disable_cgroup() {